It is a major alternative to Airbnb where you can swap houses. Registration is free, and you only pay the annual membership fee of €160 when you find your first exchange.
Subscribing to the annual membership also gives you access to Guest Points (500GP for the first subscription and 250GP for subsequent ones).
Guest Points were created to enable non-reciprocal exchanges when members cannot find someone to host them on the dates they would like to travel.
Thanks to this system, users are able to organize two types of exchanges on HomeExchange:
HomeExchange needs to understand why some of their customers leave.
Several variables are available to identify whether seniority, activity as a guest or host, sponsorship, and subscriptions taken through promotions have an impact on the churn rate.
Two datasets are provided by HomeExchange :
Datasets cover period from 2019-01-01 to 2022-11-30. As subscription are for one year, the maximal date subscription date is '2021-10-31'.
Some cleaning and data enrichment was done in SQL (DBT) :
A new table was created in SQL BigQuery
Subscription data :
loading subscription dataset.... .... cleaning stage in SQL remove : 4.37 % of data
Exchange data :
loading exchange data
Resample exchange data (monthly)
Users data :
Load Users data 414678
We will begin by examining the subscription table, which contains information for each user and their subscriptions. Each subscription is represented by a separate line in the table, and there may be multiple lines for a single user if they have multiple subscriptions.
Analyzing this table will enable us to conduct an initial investigation into the reasons why users may leave after subscribing. We can examine patterns or factors that may contribute to user churn and gain insights into which users are more likely to cancel their subscriptions.
In the following section, we will study the profile of churners using the Users table. By analyzing this table, we can delve deeper into the characteristics and behaviors of users who churn, helping us understand why they cancel their subscriptions.
As renew is indicated in the Subscription table as 1 if the user re-subscribe and 0 if not.
Therefore, the churn rate can be determined as :
CR = 1 - mean(renew)
which is equivalent to calculate the number of time a user that did not re-subscribe divided by the total number of subscribtions.
A total of 33% of churn is experience for the period 2019 - 2022, for all users.
The churn rate by continent is shown below, as well as the proportion of users that the country represents.
The churn rate for users that did 0 exchange between 2019 and 2022 is much smaller for USA's users compare to French's users, indicating that USA users have more tendency to use automatic renew.
By looking at users that subscribed for the first time after 2019 and that at least 1 exchange : - 69.4 % of USA users subscribed BEFORE doing any exchange - 61.8 % of French users subscribed BEFORE doing any exchange
We can conclude that Americans may tend to take out a subscription before finding an exchange and/or renew without even having taken advantage of their subscription.
Subscriptions are dependant on time (more subscription for example during official holidays like summer holidays for example).
Besides, data are available for 2019 to 2022, during which COVID and its lockdown strike the world. During this period, as expected, number of subscription decrease and churn rate increase.
/!\ Do not forget than there is one year between the choice of the user to renew or not and the subscription date (that's why the churn rate is shifted by one year compared to subscription count)
Some of the user that leaves during COVID can back after, shown as "return" rate in the following chart.
COVID is showing a large impact on churn rate, which can be verified with a statistical test (Fisher exact in this case).
COVID is for the period before 2021-04-15 and POST-COVID is for period after this date
| renew | 0 | 1 |
|---|---|---|
| covid_renew | ||
| 0_covid | 17595 | 27334 |
| 1_post-covid | 14015 | 36686 |
p_value: 1.23418038102e-312 Result: There is a significant difference between COVID and POST-COVID period
This impact will be keep in mind for the rest of the analysis.
As expected, the churn rate is smaller for user that did subscribe before (not a first subscription). Besides, if the first subscription was done before or during COVID, then the churn rate is larger than if it was done after COVID (after 2021-06-15), which is perfectly normal since user had more difficulty to travel
As expected, the more the users subscribed in the past, the less they leave during the following year. That means that, when the solution is working (managed to travel), user stay and continue to use it.
When a user uses a promotion for the first subscription, they have to more change to stay than if they did not use a promotion. When a user uses a promotion, but it is not their first subscription, then the behavior is reversed and the churn rate increase in case of use of promotion from 28% to 41%
**HUGE !!! HOW TO EXPLAIN THIS ???????**
Referral do not impact much the churn rate
In the previous section, we have studied subscription and users' characteristics. One other interesting feature to study is the exchange type and number as a function of time.
As expected, during look down, a lot of exchange were canceled as users could not travel. In the same way, the number of conversation and exchange finalizes during COVID period drastically drops.
However, something interesting appears on this graph: the amount of exchange increase after COVID (compare to the period before the first look down). Since the number of inscriptions (see section 3) did not increase that much after COVID, that indicate that the user that already subscribed travel more after COVID! There is alsmot 3 ( 2.97) times more exchange finalized in June 2022 compared to June 2019, while there are only 1.48 more users that subscribed (including renewals).
(Exchanges : JUNE 2021: 375 865 vs JUNE 2019: 126 410) Subscriptions : JUNE 2021 : 3 262 vs JUNE 2019 : 2 203)
**People travel more, that means that those users should be happier with the solution and could be the best ambassador for Home Exchange. It could be interesting to increase reward for users that recommended the solution to friends.**.
Based on both Subcription and Exchange tables, we want to study the comportment of users that subscribe to the solution and but not leave.
First, it is interesting to see that the number of users that registered but not subscribed is important.
Then, the impact of the number of exchanges and the typology of the users (if they are a guest or host, or both) on the churn rate will be studied.
NB: each user can leave or not several years, which give a mean churn rate by user. To calculate the "total" mean churn rate (calculated by the number of subscriptions and not by number of users), the mean churn rate will be calculated as a weighted average of the mean churn rate per user (weighted by the number of inscriptions of the user).
Almost 84% of registered users never subscribed to the solution.
It is an important number but not really surprising considering the solution offered by HomeExchange.
By looking at the number of conversations that users that never subscribed did, we can see that the large majority of register users try less than 15 conversations.
The number of conversations follows a decreasing exponential function. It is possible to estimate the number of conversations after which half of the users "disappear" (meaning that half of the initial users are not trying to find exchange after this number).
To do this, we can do a fit with a half-life functional:
Number of conversation for which the total number of users (that did not subscribed) reduce to half of its initial value is 8.9
After 9 conversations, half of the users are not trying to find exchange anymore.
It could be interesting to study in more depth why users to not subscribed, but it will not be done here as we are interested by the churn rate.
78% of users that susbcribed between 2019 and 2021 did at least one exchange, and 46% of those users did more than 5 exchanges between 2019 and 2022.
22 % of users that subscribed between 2019 and 2021 did not use the solution at all. As expected, those users have a large mean churn rate equals to 51.4% as shown on the next chart :
The use of Guest Point (guest only) does not allow to convince users to stay another year especially for first subscriber (even after COVID).
On the other hand, if users do exchange their houses, even for a first subscription, they have a much smaller churn rate!
This is the solution that is working !
The number of conversations follows a decreasing exponential function. It is possible to estimate the number of conversations after which half of the users "disappear" (meaning that half of the initial users are not trying to find exchange after this number).
To do this, we can do a fit with a half-life function:
Number of conversation for which the total number of users (that did not exchange) reduce to half of its initial value : 12.5
Users that never exchanges and did less than 12 conversations have a largest churn rate than those who did more than 12 conversations. That behavior indicated that the churner are those who do not try a lot to find an exchange.
**Maybye HomeExchange could try to motivate them with some e-mail or recommendation (for example improve quality of house picture or description, alert when users city is on the top research on the web site etc...)**